Cette page appartient aux archives web de l'EPFL et n'est plus tenue à jour.
This page belongs to EPFL's web archive and is no longer updated.

TCP/IP

date/time of the exam
Salut,
short question because i didnt find any information about this on the page.
at what time is the exam?
Monday 5th @ ??

thx
Philipp
Posted by Philipp Stumpf at 17:51
Comments (2)
appli.ppt / ipagain.ppt
Hi, I, again, have some questions:

appli.ppt
p.9
If I'm not mistaken, a TCP server port is a port on which a machine listen for incoming connections. So, In passive-mode, why is port 1515 on S not listed as such ? And, always in passive-mode, the OK arrow of the second connection is reversed, I think.

p.13, 3.
It is said that a root name server knows the IP addresses of all level-2 domains. But there's a lot of level-2 domains ! So is it not all top-level domains ? And then, another name server which knows all level-2 domains related to this top-level domain gives the final answer ?
And how is the database of these top-level (level-2) domains maintained up-to-date at a root server ?


ipagain.ppt
p.4
R1 sends an ICMP redirect to inform A that a shorter route exists through R2. Ok, but how does R1 knows about the connection A-R2 ? R1 being on link with A and R2 doesn't automatically imply that A and R2 are also on link. So, does the ICMP redirect function only with certain network topologies ?

p.14/43
When PMTU is enabled, isn't the "don't fragment" flag set ? Is this case, a router cannot fragment a TCP packet even if the PMTU estimation failed, all it can do is send an ICMP message to inform that the packet is too big.

p.22/44
In this slide, the physical address seems to be deducted from the IP address, but where does this 33:33 comes form ? Is it just a default value set when some part of the physical address can't be deducted.
The solicited-node address ff02::1:80b2:9c26/128 corresponds to a machine which has a physical address ending by 80.b2.9c.26. This doesn't correspond to lrcsun13' physical address but it does correspond to the translated IP address of lrcsun13. So, will lrcsun13 reply to ff02::1:80b2:9c26/128 ?

Ok, it's all for tonight.
Posted by René Giller at 15:50
Comments (3)
tp4: question on cwnd (page 10), if you can not solve it how to go on with tp
we give you a lot of hints for this question (see blog), but if still you can not find what limits cwnd at value 91 (or 92), in order to do the rest of the tp successfully you can just assume for this question that:

- cwnd is limited by value 91 (or 92) and can not go above this value even if congestion control mechanisms and very small losses or no-losses would let cwnd goes above this value
-RTT is not cause for cwnd limitation, but RTT is a consequence - in experiment 1 and 2! In experiment 4 it will not be like this!

Also be aware (we already told this to you) that in this tp system and experiments main RTT component comes from queuing, other components are almost negligable.
Posted by Slavisa Sarafijanovic at 8:30
Research Exos are Corrected
Most of you who handed in the research exercise did a good job, congratulations. However, the parts we considered most important (captive portal, overall solution) were generally answered with less care than the other questions. How we graded: if you answered most of the introduction modules well you got a bonus of 0.25. if you additionally explained the captive portal and the overall epfl solution very well, you got the full bonus of 0.5. We then ranked the people who got the full bonus and proposed to the best ones to make the presentation which can give an additional bonus of 0.5.

The bonuses are on display in front of bc260 (as usual, bring your sciper number along).
We also invite you to come and see us to get more feedback on your report.
Posted by Manuel Flury at 13:47
tp4 experiment 1
Hello,
We've just tried to apply the command "echo 0 > /proc/sys/net/ièv4/tcp_bic" (p.9) but the file tcp_bic wasn't present in the system (of the Workstation). Did we miss something? Can we just pass through it and continue the TP?
Thanks in advance,
The Quang.
Posted by The Quang Nguyen at 11:04
Comments (2)
tp4 experiment 4
Hello,

We tried to run after_restart.sh as written in the tp4 document, but there is no /proc/sys/net/ipv4/tcp_congestion_control on any of computers we tried. What should we do?

Aleksandar
Posted by Aleksandar Dragojevic at 21:47
Comments (2)
max-min fairness
Hi, there's something I didn't get about max-min fairness :

cc.pdf, p.9 third paragraph of Network Model :
... The name "max-min" comes from the idea that is is forbidden to decrease the share of sources that have small values, thus, in some sense, we give priority to flows with small values.

But by definition of max-min fairness, an increase of any rate is a the cost of a decrease of an already smaller one.

To me, theses two sentences seem contradictory and from the definition, I don't see why it would be forbidden to decrease a small rate. I agree that it would not be fair to further decrease an already small rate, but I don't see anything in the definition that would prevent it.
Posted by René Giller at 17:37
Comments (5)
typos in cc.pdf
Hi, I think I found some typos in cc.pdf :

p.6, first paragraph :
...Source i enters node i, uses links [(i + 1) mod I] and [(i + 2) mod I], ...

but this is not consistent with Figure 1.2 below, it should be :
...Source i enters node i, uses links [i mod I] and [(i + 1) mod I], ...

the same applies 2 lines below.


p.19 last line :
...then we must have u0 = 0.

but it should be
...then we must have u0 = 1.

otherwise, it won't be an increase anymore.


p.26 third point of the list :
additive increase, multiplicative principle is used

the word decrease is missing after multiplicative
Posted by René Giller at 17:17
Comments (1)
Research exercise
Hi,

When will be the office hour to hand in research exercise?

Ghazale Hosseinabadi
Posted by Ghazale Hosseinabadi at 19:52
Comments (1)
tp4: more hints for question on cwnd (page 10)
Let L be the list of things we (already) proposed to you to look at:

L={sending window, cwnd, offered window, scale option, in_flight variable}

Then:

1. ALL elements from L are important for the explanation

2. a good way to start is to express each of elements from L as function of all other elements from L on which it is dependand directly (as explained in the lectures and tp document itself); then look what are all the things that can constrain cwnd.

3. then have in mind the NUMERIC VALUE of cwnd that you observe in the experiment (printed by tcpsender) when tcp flow is stabilized, find other numeric values of elements from L that could constrain cwnd, and verify which of them actually do constrain cwnd (i.e which constraint is the most critical, i.e. which one is in action).
Posted by Slavisa Sarafijanovic at 17:56
Page : « Previous 1 2 3 4 5 6 7 8 9 10 11 12 Next »